🌙 Sandman Tales v2

AI-Powered Multilingual Bedtime Stories

Team ClawCutters — Mistral Worldwide Hackathon 2026

Technical 20% Creativity 20% Usefulness 20% Demo 20% Track Alignment 20%
🎯 Usefulness

The Quiet Crisis

27.6%
of Australians were born overseas — ABS Census

Millions of kids grow up hearing a second language at home.
But that language fades as English takes over at school.

"I don't want Sophie to forget French. It's who we are." — Sophie's mum
💡 Creativity

Sandman Tales

Bedtime stories that keep heritage languages alive


🎤
Parent speaks
in English
→
🕷️
AI writes story
in child's heritage language
→
🔊
Child listens
to personalised bedtime tale

10 languages · Personalised with child's name · Watercolour illustrations · Three audio layers

💡 Creativity

Meet the Team

Six AI agents inspired by cultural mythology

🌳 Papa Bois
Orchestrator

🕷️ Anansi
Storyteller

🗣️ Ogma
Language Guardian

🙏 Devi
Voice & Audio

🦆 Firefly
Builder

🐭 Jerry
Concierge

⚙️ Technical

System Architecture

%%{init: {"theme": "dark", "themeVariables": {"primaryColor": "#a78bfa", "primaryTextColor": "#e2e8f0", "primaryBorderColor": "#818cf8", "lineColor": "#f59e0b", "secondaryColor": "#1e1b4b", "tertiaryColor": "#0f172a", "fontSize": "14px"}}}%% graph TB subgraph CLIENT["🖥️ Frontend — React + Vite"] UI["Story Creator
Story Player
Story Library
Voice Picker"] end subgraph BACKEND["⚡ Backend — FastAPI on Railway"] API["REST API + WebSocket"] CACHE["Prompt Cache
SHA-256 Dedup"] ORCH["🌳 Papa Bois
Orchestrator"] end subgraph MISTRAL["🤖 Mistral AI"] AGENTS_API["Agents API
Conversations + Handoffs"] LARGE["Mistral Large 3
Story Generation"] VOX["Voxtral
Speech-to-Text"] end subgraph ELEVEN["🔊 ElevenLabs"] STT["Scribe v1 — STT"] TTS["Multilingual v2 — TTS"] STREAM["WebSocket Streaming"] SFX["Sound Effects"] MUSIC["Music Compose"] VOICES["Voices API"] EA["ElevenAgents"] end subgraph STORAGE["💾 Storage"] TURSO["Turso
SQLite Cloud
Tokyo Region"] IMAGEN["Google Gemini Image Generation
Illustrations"] end UI -->|HTTPS| API API --> CACHE CACHE -->|cache miss| ORCH ORCH -->|delegates| AGENTS_API AGENTS_API --> LARGE ORCH -->|dual STT| VOX ORCH -->|dual STT| STT ORCH -->|narrate| TTS ORCH -->|stream| STREAM ORCH -->|ambient| SFX ORCH -->|lullaby| MUSIC ORCH -->|browse| VOICES ORCH -->|chat| EA ORCH -->|illustrate| IMAGEN API -->|persist| TURSO CACHE -->|dedup| TURSO style CLIENT fill:#1e1b4b,stroke:#818cf8,color:#e2e8f0 style BACKEND fill:#1e1b4b,stroke:#f59e0b,color:#e2e8f0 style MISTRAL fill:#2d1f5e,stroke:#a78bfa,color:#e2e8f0 style ELEVEN fill:#1b2e1b,stroke:#34d399,color:#e2e8f0 style STORAGE fill:#2e1f1b,stroke:#f59e0b,color:#e2e8f0
⚙️ Technical

Agent Pipeline — Story Creation Flow

%%{init: {"theme": "dark", "themeVariables": {"primaryColor": "#a78bfa", "primaryTextColor": "#e2e8f0", "lineColor": "#f59e0b", "fontSize": "13px"}}}%% sequenceDiagram participant P as 🎤 Parent participant O as 🗣️ Ogma
Language Guardian participant C as 📋 Cache participant PB as 🌳 Papa Bois
Orchestrator participant G as 🛡️ Guardrail participant A as 🕷️ Anansi
Storyteller participant D as 🙏 Devi
Voice/Audio participant F as 🦆 Firefly
Builder P->>O: Voice input (mic) activate O O->>O: ElevenLabs STT (Scribe v1) O->>O: Mistral Voxtral STT O->>O: Cross-validate + detect language O-->>C: Transcribed text + language deactivate O C->>C: SHA-256 hash check alt Cache Hit C-->>P: Return cached story else Cache Miss C->>PB: Forward to orchestrator activate PB PB->>G: Content safety check G-->>PB: ✅ Safe PB->>A: Delegate via Agents API activate A A->>A: Mistral Large 3 generates story A->>A: Split into 6 scenes A-->>PB: Story JSON (6 scenes) deactivate A par Audio Generation PB->>D: Generate narration (TTS) PB->>D: Generate SFX per scene PB->>D: Compose lullaby and Illustration PB->>F: Generate 6 scene images F->>F: Gemini Image Generation watercolours end D-->>PB: 🔊 Audio (narration + SFX + lullaby) F-->>PB: 🎨 6 illustrations PB-->>C: Store in Turso deactivate PB C-->>P: Complete story with media end
⚙️ Technical🏆 Track Alignment

Mistral Agents API — Orchestration Detail

%%{init: {"theme": "dark", "themeVariables": {"primaryColor": "#f59e0b", "primaryTextColor": "#e2e8f0", "lineColor": "#a78bfa", "fontSize": "13px"}}}%% graph TB subgraph PAPA["🌳 Papa Bois — Orchestrator Agent"] direction TB CONV["Conversations API
Multi-turn context"] HAND["Handoffs
Agent delegation"] TOOLS["Function Calling
4 ElevenLabs tools bound"] end subgraph ANANSI["🕷️ Anansi — Storyteller Agent"] direction TB ML3["Mistral Large 3"] JSON["JSON Mode
Structured scene output"] CULT["Cultural sensitivity
Language-aware prompts"] end subgraph DEVI_TOOLS["🙏 Devi — Bound Function Tools"] TTS_F["generate_tts()"] SFX_F["generate_sound_effect()"] LUL_F["compose_lullaby()"] SEARCH["search_cultural_context()
via Tavily"] end subgraph OGMA_DUAL["🗣️ Ogma — Dual STT"] EL_STT["ElevenLabs Scribe v1"] VOX_STT["Mistral Voxtral"] CROSS["Cross-validation
Best-of-two"] end PAPA -->|handoff| ANANSI PAPA -->|function call| DEVI_TOOLS OGMA_DUAL -->|transcription| PAPA style PAPA fill:#44337a,stroke:#f59e0b,color:#e2e8f0 style ANANSI fill:#44337a,stroke:#a78bfa,color:#e2e8f0 style DEVI_TOOLS fill:#1b3a2a,stroke:#34d399,color:#e2e8f0 style OGMA_DUAL fill:#1b2e4a,stroke:#818cf8,color:#e2e8f0

4 pre-registered Mistral Agents · Conversations for multi-turn · Handoffs for delegation · Function calling for ElevenLabs tools

🏆 Track Alignment

ElevenLabs — All 7 Tools in Architecture

%%{init: {"theme": "dark", "themeVariables": {"primaryColor": "#34d399", "primaryTextColor": "#e2e8f0", "lineColor": "#f59e0b", "fontSize": "13px"}}}%% graph LR subgraph INPUT["Input Layer"] MIC["🎤 Parent Mic"] CHAT["💬 Text Chat"] end subgraph STT_LAYER["① ② Speech-to-Text"] S1["① ElevenAgents
Jerry Concierge
Conversational"] S2["② Scribe v1
STT Transcription"] end subgraph VOICE_LAYER["③ Voice Selection"] S3["③ Voices/Get API
Browse & Preview"] end subgraph AUDIO_LAYER["④ ⑤ ⑥ ⑦ Audio Generation"] S4["④ TTS Batch
Multilingual v2
10 languages"] S5["⑤ TTS WebSocket
Real-time Streaming"] S6["⑥ Sound Effects
Contextual SFX"] S7["⑦ Music Compose
Lullaby Generation"] end subgraph OUTPUT["🔊 3-Layer Audio"] NAR["Narration"] AMB["Ambience"] MUS["Music Bed"] end MIC --> S2 CHAT --> S1 S1 --> S3 S2 --> S3 S3 --> S4 S3 --> S5 S4 --> NAR S5 --> NAR S6 --> AMB S7 --> MUS style INPUT fill:#1e1b4b,stroke:#818cf8,color:#e2e8f0 style STT_LAYER fill:#1b3a2a,stroke:#34d399,color:#e2e8f0 style VOICE_LAYER fill:#1b3a2a,stroke:#34d399,color:#e2e8f0 style AUDIO_LAYER fill:#1b3a2a,stroke:#34d399,color:#e2e8f0 style OUTPUT fill:#2e1f1b,stroke:#f59e0b,color:#e2e8f0
🎪 Demo

Live App Demo

Login

Starfield animation. Pre-seeded test accounts in Turso (SQLite cloud, Tokyo region).

PBKDF2-SHA256 password hashing.

🔗 Live: sandmantales-api-production.up.railway.app

🎪 Demo

Create a Story

Sophie's Story

🎤 Mic button → Ogma dual-STT
(ElevenLabs + Mistral Voxtral)

🌍 10 languages in dropdown

👧 Child's name woven into every scene

📝 Parent describes the day in English → story generated in French

⚙️ Technical🎪 Demo

Pipeline in Action

9 Agents, Real-Time

🗣️ Ogma transcribes → ✓

📋 Cache checks SHA-256 → ✓

🌳 Papa Bois orchestrates via Mistral Agents API → ✓

🛡️ Guardrail validates → ✓

🕷️ Anansi writes story with Mistral Large → running...

🙏 Devi: TTS + SFX + Lullaby → pending

🦆 Firefly: Illustrations → pending

⚙️ Technical

Pipeline Complete

All Green ✅

Every node complete. Story, illustrations, three audio layers generated.


Cache layer: repeat prompts cost zero. SHA-256 dedup on every story + audio clip in Turso.

🎪 Demo💡 Creativity

Sophie's Story

🇫🇷 Sophie et la Baleine de Nuages

📖 Six scenes in French

🎨 Watercolour illustrations (Gemini Image Generation)

🔊 Three audio layers:

  • Narration — ElevenLabs Multilingual v2
  • Sound effects — ElevenLabs SFX
  • Lullaby — ElevenLabs Music Compose

🎧 10 languages supported

🎪 Demo

Story Library

18 Stories, 10 Languages

🇬🇧 🇫🇷 🇯🇵 🇮🇳 🇪🇸 🇵🇹 🇩🇪 🇨🇳 🇸🇦 🇰🇷

Every story personalised with the child's name and world.


Every night Sophie listens, the stories grow with her. That's not a feature — it's a relationship.

🏆 Track Alignment

All 7 ElevenLabs Tools

1. ElevenAgents

Jerry — conversational story concierge, helps parents pick voices and themes

2. Speech-to-Text

Ogma — dual STT alongside Voxtral for cross-validation

3. Voices/Get

Voice browser — parents preview narration voices before generating

4. Text-to-Speech

Devi — batch narration across 6 scenes, 10 languages

5. TTS WebSocket

Real-time streaming for scene-by-scene playback

6. Sound Effects

Contextual SFX per scene — ocean waves, birdsong, rain

7. Music Compose

Lullaby generation — gentle music bed under narration
🏆 Track Alignment

Mistral Integration

Mistral Agents API

Papa Bois orchestrates agent delegation via Conversations + Handoffs

Mistral Large 3

Anansi — story generation, scene splitting, cultural sensitivity

Voxtral

Ogma — speech-to-text alongside ElevenLabs for dual transcription

Vibe CLI

Firefly — built frontend components, Papa Bois planned architecture

%%{init: {"theme": "dark", "themeVariables": {"primaryColor": "#f59e0b", "primaryTextColor": "#e2e8f0", "lineColor": "#a78bfa"}}}%% graph LR V["🎤 Voxtral STT"] --> PB["🌳 Papa Bois
Agents API"] PB --> AN["🕷️ Anansi
Mistral Large"] PB --> VB["🔧 Vibe CLI
Firefly"] style PB fill:#44337a,stroke:#f59e0b style AN fill:#44337a,stroke:#a78bfa
⚙️ Technical

Engineering Maturity

🎨 LoRA Fine-Tuning: Trained Stable Diffusion on 20 storybook images (Gemini Image Generation generated). 500 steps, 15 minutes, 3.1MB weights.


🧑‍⚖️ LLM-as-Judge: Pixtral evaluated base vs LoRA side-by-side.
Base SD 1.5: 6.8/10 → LoRA: 9.0/10 → Gemini Image Generation: 9.5/10


📊 Decision: We shipped Imagen. Knowing when NOT to use your own model is the innovation.


💾 Prompt Cache: SHA-256 dedup. Zero-cost repeat requests. All stored in Turso cloud DB.


✅ 33/33 smoke tests passing — auth, CRUD, cached audio, story gen, all ElevenLabs endpoints, Mistral agents

🎯 Usefulness

Market Opportunity

50M
multilingual households
worldwide
29
languages
supported
$9.99
/month
freemium

Every night Sophie listens, the stories grow with her.
Personalisation data builds a relationship competitors can't copy overnight.

Team ClawCutters

Six mythology-inspired AI agents · Seven ElevenLabs tools · Mistral Agents API
LoRA fine-tuning · Pixtral judge · Turso cloud · Built in 48 hours


Bonne nuit, Sophie 🌙

🔗 sandmantales-api-production.up.railway.app